Case study · Nimbus, a company the team generated for demos
The apparent success
New revenue rose 80%. How much of it is staying?
Nimbus’s New MRR, the monthly revenue its new subscriptions bring, rose from the first half of to the second. Good news, on the face of it. So we asked what the growth is made of, and how much of it stays.
Where it came from
Most of it came from one plan: Enterprise.
Split by the plan each subscription was sold on, Enterprise brought of the rise.
So the growth rests mostly on one plan.
The contradiction
But the revenue lost to churn grew every quarter.
Churned revenue rose from in the first quarter of to in the last of .
Is the growth leaking out?
Against new revenue, it fell.
For every $100 of new revenue, churn took in the first half of and in the second. Net new revenue more than doubled.
The discovery
Customers aren’t leaving faster.
Cohort Analysis follows each month’s new accounts: are active in their first month on average, about half the next, about a fifth a year later. The newer months follow the same curve.
So churn grows because there are more customers, not because they leave sooner.
The check
It holds, and the churn is counted right.
New MRR’s 95% interval runs from to , all of it above no change.
- Checked against an independent count. Churned revenue over the two years, , matches an independent count of the same data.
- Not Deep Analysis’ churn. Its “Churned MRR” counts every change in revenue, not only churn, so it isn’t used here.
- “Active” means a session that month, not revenue.
- Not a cliff. The product’s AI read the newest months as a "severe retention cliff": those months haven’t happened yet.
And a person approved what New MRR counts.
On Nimbus’s Living Map, New MRR sits under Net New MRR, and a person approved what it means: the run used that definition. Of the map’s metrics, only carry that check.
What we’d decide
Proposed decision
Keep investing behind the growth, and judge churn against new revenue, not in dollars.
In Nimbus’s data, which the team generated for demos, churn grew with the business, not faster than it.
Next: Report churn for every $100 of new revenue, and split it by plan before moving acquisition.
Provisional · further evidence required
The strategy, from what we know and what we don’t
Report churn against new revenue, not in dollars.
Because: Churn in dollars will keep rising as the business grows; for every $100 of new revenue it fell from to .
Not known: Whether that holds as Enterprise accounts age.
Work on the second month.
Because: About half of each month’s accounts are active in their second month, against in the first, and the curve is the same for every month.
Not known: Why they go quiet. The data shows sessions, not reasons.
Split churn by plan before moving acquisition.
Because: Over the two years Enterprise and Pro lost about the same revenue to churn ( and ), Starter .
Not known: Each plan’s revenue base, so which loses the larger share of what it brings.
What would change it: If churn for every $100 of new revenue climbs back over , or a newer month’s accounts fall below the curve.
The full analysis the definition, the plans, new and lost revenue, the cohorts, what ran, the checks, the queries
The definition
New MRR: SUM(mrr_delta) on fct_subscriptions, where type = 'new'. Approved by a person; the run used it.
On Nimbus’s Living Map, New MRR hangs under Net New MRR. Of the map’s metrics, a person has approved ; New MRR is one. Churned MRR, beside it, was matched by the machine and no person has reviewed it; Churn Rate was rejected by a person.
The rise, plan by plan
| Plan sold | Change a quarter | 95% interval | On the first half |
|---|---|---|---|
| Enterprise | |||
| Pro | |||
| Starter |
Shares of the rise: Enterprise , Pro , Starter . The three add up to the rise, and every part’s interval stays above zero.
New and lost revenue, each quarter
New MRR from its saved run; churned revenue from KPI Drill-Down; for every $100, our arithmetic.
| Quarter | New revenue | Lost to churn | Lost for every $100 gained |
|---|
The halves of : new revenue and a quarter; lost and ; net and . Two-year totals: new , lost , each matching an independent count of the same data.
Lost to churn, by the plan it left
Over the two years, from KPI Drill-Down.
Enterprise , Pro , Starter .
The cohorts
Accounts by the month they signed up; active means a session that month.
On average across the months that have it, of a month’s accounts are active in their first month, in the next, in the third and twelve months on (the months of ). The grid leaves out every month that hasn’t happened: the product returns 0 there, and its AI read those zeros as a fall.
What ran
New MRR: KPI Drill-Down ( s), then Deep Analysis in Trend mode ( s), by quarter, January 2024 to December 2025: rows, saved on 27 September 2026. Churned MRR: KPI Drill-Down, 11 November 2026. Cohort Analysis, accounts by signup month: 11 November 2026, on the app with fixes not yet released. How they work, on Product.
The checks
Why the churn figures are the right ones.
Deep Analysis’ “Churned MRR” sums every change in subscription revenue, new and upgrades included, because the definition it found under that name has no churn filter; its quarters don’t match Drill-Down’s, so it isn’t used. Drill-Down keeps the filter (type = 'churn'), and its two-year total matches an independent count of the same data, as New MRR’s does.
What the product’s AI wrote, and why it isn’t evidence
- ""
- ""
- ""
And on the cohorts: "" Written by the product’s AI after the numbers and quoted as it wrote them: none of it is evidence. Its share of the growth counts each account’s plan; the above counts the plan each subscription was sold on.
Who decided what
A person approved the definition New MRR’s run used. The decision above is proposed, not taken. Nobody has acted on any of it: Nimbus is a company the team generated for demos, and none of this is a customer’s result.
About the data
Nimbus is a company the team generated for demos, so what holds here holds in its data, not in a market. Every figure is the product’s own output, saved by its runs, except the dollars lost for every $100 gained and the cohorts’ averages, which we worked out from them. How we check a case.
The queries behind the numbers
The engine wrote these from the definitions. No AI touches them; the same question gives the same query.
, … the joins for each factor follow: lines in all
, …
Design partners
Bring your own question.
Design partners take the question their teams keep asking through their own map, with us.
Become a design partner